pytorch-transformers/pytorch_transformers/modeling_transfo_xl_utilities.py [121:144]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        else:
            # construct weights and biases
            weights, biases = [], []
            for i in range(len(self.cutoffs)):
                if self.div_val == 1:
                    l_idx, r_idx = self.cutoff_ends[i], self.cutoff_ends[i + 1]
                    weight_i = self.out_layers[0].weight[l_idx:r_idx]
                    bias_i = self.out_layers[0].bias[l_idx:r_idx]
                else:
                    weight_i = self.out_layers[i].weight
                    bias_i = self.out_layers[i].bias

                if i == 0:
                    weight_i = torch.cat(
                        [weight_i, self.cluster_weight], dim=0)
                    bias_i = torch.cat(
                        [bias_i, self.cluster_bias], dim=0)

                weights.append(weight_i)
                biases.append(bias_i)

            head_weight, head_bias, head_proj = weights[0], biases[0], self.out_projs[0]

            head_logit = self._compute_logit(hidden, head_weight, head_bias, head_proj)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



pytorch-transformers/pytorch_transformers/modeling_transfo_xl_utilities.py [215:237]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        else:
            # construct weights and biases
            weights, biases = [], []
            for i in range(len(self.cutoffs)):
                if self.div_val == 1:
                    l_idx, r_idx = self.cutoff_ends[i], self.cutoff_ends[i + 1]
                    weight_i = self.out_layers[0].weight[l_idx:r_idx]
                    bias_i = self.out_layers[0].bias[l_idx:r_idx]
                else:
                    weight_i = self.out_layers[i].weight
                    bias_i = self.out_layers[i].bias

                if i == 0:
                    weight_i = torch.cat(
                        [weight_i, self.cluster_weight], dim=0)
                    bias_i = torch.cat(
                        [bias_i, self.cluster_bias], dim=0)

                weights.append(weight_i)
                biases.append(bias_i)

            head_weight, head_bias, head_proj = weights[0], biases[0], self.out_projs[0]
            head_logit = self._compute_logit(hidden, head_weight, head_bias, head_proj)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -